Exposure to Hazards Assessment based on ‘POP-to- GUF ... · [Final Draft] Aug 2018 Prepared for Expert Group on Disaster-related Statistics for Asia and the Pacific1 Introduction:

1

Exposure to Hazards Assessment based on ‘POP-to-

GUF’ Methodology

Sequence of Steps

An Operational Manual using QGIS

[Final Draft] Aug 2018

Prepared for Expert Group on Disaster-related Statistics for Asia and the Pacific1

Introduction:

This manual provides step-by-step instructions for implementing a methodology for calculating

high resolution gridded-population and urban (built-up area) maps and statistics in GIS for

overlay with the available maps of hazards (e.g. potential flood inundation maps) for estimating

population exposure to hazards – one of the core components of disaster risk measurement.

The methodology is implemented using freely available open source GIS software tools, in

particular QGIS and related applications or plugins, most notably SAGA tools. Increasingly,

computations and production of statistics in GIS can be expedited by write code for the

conversions or calculations using Python script. Working with GIS can involve a need for tests or

trouble-shooting and there is often more than one way to accomplish the same function. This

manual provides the step-by-step instructions for operating within the QGIS or SAGA platforms.

For this manual, we utilize a real-life example and real data for Thailand. The estimates are

calculated based on the official national population census data (counts of population by most

detailed available geographic areas) along with geospatial data, particularly the Global Urban

Footprint (GUF) processed and made available by the German Space Agency (DLR), based on

radar satellite imagery. Other relevant geospatial data inputs utilized include a map of land

cover types, which was produced by the European Space Agency (ESA) Climate Change

Initiative, and hazard area maps accessible online. The methodology integrates information

from the population census into the GUF gridded data to estimate population density by grid

cell, thus in short-hand it is called the ‘Pop-to-GUF’ method.

For this manual, QGIS version 2.14 is recommended (more stable than the 2.18 and 3.03 recent

releases). The package includes GDAL, GRASS and SAGA processing modules. In some cases, for

convenience or efficiency, calculations be done with the native SAGAGis 6.3.0 and re-imported

to QGIS or similarly with any other GIS package using a raster format read by QGIS. Thus, it’s

recommended to also download the SAGA open source package (http://www.saga-

gis.org/en/index.html), which is inter-operable with the QGIS.

1 This manual developed by Jean-Louis Weber (Consultant) with Daniel Clarke (UNESCAP), Trevor Clifford

(Consultant), and Gao Xian Peh (Consultant), with valuable feedback from the Disaster Management Agency of

Indonesia (BNPB), the Dept. of Disaster Prevention and Mitigation (DDPM) of Thailand and the National Statistical

Office (NSO) of Thailand.

2

Table of Contents

1 Collect data .................................................................................................................... 4

1.1 Create a folder and name it Downloaded Data ......................................................... 4

1.2 Statistical Tables ...................................................................................................... 4

1.3 Vector files (shapefiles) ............................................................................................ 5

1.4 Raster files ............................................................................................................... 6

2 Upload layers to QGIS and prepare downloaded data ..................................................... 7

2.1 Upload the national boundaries shapefile (ADM0) ................................................... 7

2.2 Upload and pre-process GUF Data ............................................................................ 8

2.2.1 Merge GUF tiles into one layer .......................................................................... 8

Use QGIS/Raster/Miscellaneous/Merge ......................................................................... 8

2.2.2 Clip out unused GUF data .................................................................................. 9

2.3 Upload and pre-process land cover and hazard maps ............................................ 10

2.3.1 Extract a working window from the global ESACCI land cover and hazad maps 10

2.3.2 Clip out unused raster data ............................................................................. 11

2.4 Save the project ..................................................................................................... 12

3 Prepare Shapefiles for Assimilation of Population Data ................................................ 12

3.1 Define project projection system (CRS: Coordinates Reference System) .................. 12

3.1.1 Set the CRS for the project .............................................................................. 13

3.2 Upload the national boundary (ADM0) shapefile to QGIS ....................................... 13

3.2.1 Upload the national boundary (ADM0) shapefile to QGIS ................................ 13

3.2.2 Check the CRS of ADM0 in the Properties/General box .................................... 13

3.3 Project (or re-project) ADM0 to the project CRS ..................................................... 14

Select the shapefile layer (ADMO) and Save As... .......................................................... 14

3.3.1 Verify the new coordinate reference system (CRS) for ADMO .......................... 14

3.4 Project (or re-project) other Shapefiles (as needed) ............................................... 15

3.4.1 Upload the ADM1, ADM2, ADM3 shapefiles to THA_v1 ................................... 15

4 Create a raster mask layer with properties for data assimilation ................................... 15

4.1 Creation of the 1_No Data Mask with 100m pixels ................................................. 16

4.2 Creation of the 0_No Data Mask with 100m pixels ................................................. 17

5 Assimilation of Input data ............................................................................................ 18

3

5.1 Prepare Population Census tables .......................................................................... 18

5.2 Introduce Population Census statistics into the project shapefile ............................ 18

5.2.1 Upload CityPop2000_2010_Districts to the project. ......................................... 18

5.2.2 Join to the administrative region shapefile ...................................................... 18

5.3 (Re)Project and Resample raster layers .................................................................. 19

5.3.1 Re-project (convert, warp...) each of the raster layers to EPSG:3395 ................ 19

5.3.2 Remove the old (prev. projection) files from the THA_v1 project ..................... 19

5.4 Resample raster data to common grid .................................................................... 20

5.4.1 Resample GUF data to 100x100m grid ............................................................. 20

5.4.2 Resample ESACCI land cover data 100x100m grid ............................................ 20

5.4.3 Resample Hazard/Risk data (Flood_Risk_UNEP_GRID3395.tif) ......................... 22

6 Weighting GUF pixels according to their agglomeration ................................................ 23

6.1 Clip-out the sea and convert GUF pixels to 1 value ................................................ 25

6.2 Smoothing Clipped GUF data.................................................................................. 26

6.3 Clip out smoothed values outside of GUF cells (optional)........................................ 28

7 Estimate population living out of GUF pixels ................................................................. 29

7.1 Define area where dispersed population is expected to live outside of GUF ............ 30

7.1.1 Selection of classes of the ESACCI land cover maps .......................................... 30

7.1.2 Elimination of dispersed population overlay (duplication) with GUF pixels ...... 32

7.2 Estimation of population in Non-GUF pixels ........................................................... 33

7.3 Extract values from pixels to Administrative Regions (e.g. districts) ........................ 33

7.4 Subtract estimated Non-GUF population from ....................................................... 34

7.5 Check results and Fix anomalies ............................................................................. 35

7.6 Calculate population density per GGUF point ......................................................... 37

8 Produce final gridded population density maps ............................................................ 38

8.1 Rasterisation of GGUF points’ population density ................................................... 38

8.2 Calculation of final POPtoGUF Rasterized Results ................................................... 39

9 Estimation of population exposure to flood hazard areas ............................................. 40

9.1 Creation of two risk intensity levels: Low risk and High risk .................................... 40

9.2 Calculation of population in Low risk and High risk exposure areas ......................... 40

9.3 Extraction of statistics ............................................................................................ 42

9.4 Presentation of results ........................................................................................... 44

4

1 Collect data

This first section of the Operational Manual summarizes the basics on the datasets, by file type,

that will need to be downloaded and organized in your project folder.

1.1 Create a folder and name it Downloaded Data

1.2 Statistical Tables

Tables are collected for Population census statistics. Optional is to collect tables on other topics

relevant to assessing exposure or vulnerability to hazards, where available, such as crop

production statistics, for example. Statistical tables can be integrated into GIS formats if they

contain a reference to a place or geographic location, such as government administrative

regions (e.g. districts).

Tables can be handled and processed with different software packages:

• Spreadsheets: e.g. MSExcel (.xls and .xlsx), LibreOffice or OpenOffice (.ods),

• Data Base Management Systems: e.g. POSTGRESQL/POSTGIS, MSAccess, SQLite/Spatialite.

Common tables formats are .dbf (Note: .dbf headings of 8 digits maximum), .csv ot .txt

• GIS attribute tables to shapefiles (and some grid file formats); they can be viewed with

spreadsheets as .dbf or .csv file (Note: MSExcel cannot save tables as .dbf; LibreOffice and

Open Office can).

The pollution statistics tables are downloaded in spreadsheet format (.csv or .xlsx). Cases exist

where detailed statistic are supplied in .pdf format only. It is possible to convert .pdf to .xls

or .csv but it requires careful and long check of the tables as well as elimination of parasitic

graphic features.

QGIS reads and saves tables in .dbf fomat. Conversions to .dbf format is done, in principle using

Save As... or Import/Export functions within QGIS, but not all packages can read all formats.

Tables used by DBMS and GIS calculations record:

In rows: shapes, features, statistical units (e.g. administrative divisions), objects...

In columns: fields, attributes (e.g. name, codes, variables’ values)

5

IMPORTANT : In rows, shapes’ internal IDs are

automatically given by the GIS when the shapefile is

created. It corresponds to the order of the rows in

the initial shapefile. It means that this order should

NEVER be modified. GIS package manage this issue.

Spreadsheets don’t visualize the internal ID. As a

security, it is recommended upon importing table

into QGIS to first create a new field named ROW_ID

using QGIS vector calculator, and requesting @row

number.

The statistical tables are joined with shapefiles in GIS only if the imported statistics and

shapefile have common identifiers, ideally common administrative codes (geographic codes

identifying location for the relevant values). Often coding or naming of geographic areas vary,

so careful consistency checks are needed to join the statistics into your GIS project.

From the population census, collect statistics for the most detailed administrative divisions

available (usually ADM2 or ADM 3) and the hierarchy of levels. ADM0 stands for the whole

country. ADM1 usually stand for regions and ADM2 for districts, but the terminology and

division patterns vary from country to country. More detailed levels vary even more from

country to country. Statistics on Population Censuses by ADM0, ADM1 and ADM2 are available

on international websites (e.g. CityPop, World Bank). National statistical offices increasingly

provide open access to population census data, including that data at the most detailed level

(e.g. villages), with geographic names and codes to use as common identifiers for linking with

shapefiles.

1.3 Vector files (shapefiles)

Vector files record points, lines (e.g. roads) and polygons (e.g. administrative divisions) as a

drawing to which is attached an attribute table. The commonly used format for vector files is

ESRI shapefile (.shp). It is a de facto standard for ArcGis, QGis, SAGAGis. There are other

formats used such as for example in MapInfo (.mif), AutoCAD (DWG) or Google Earth (kml,

kmz). Conversion between vector files format is done by the various GIS packages.

The shapefiles to download for POPtoGUF methodology are administrative divisions. They can

be obtained through mapping agencies, international websites such as GDAM, GitHub, or

OpenStreetMap or directly from the National Statistical Office or other responsible government

agencies. Use of government official shapefiles is preferable for best consistency between

shapes and tables (common ADM codes). When shapefiles and tables come from two different

sources, careful check of codes is necessary.

6

Other shapefiles used in our assessment are hazard perimeters or zones (unless in raster

format). There are usually national data available based on modelled probabilistic assessments

of likelihood of hazards under various scenarios. The national maps of hazard areas are an

important resource for disaster reduction, often produced by mapping, environmental, or

disaster management agencies at the national level. These resources should be shared with

other agencies, e.g. national statistics offices, for statistical purposes. International sources

also exist, including hazard area vector or raster files, at the global scale. Unfortunately,

currently, the global assessments are not yet available at high levels of resolution for use at

national or lower scale assessments, except in pilot or sample exercises like the one used in

this manual.

High resolution land cover maps are also needed Land cover maps are usually available as

shapefiles when produced at the national level or regionally in the case of ICIMOD, for example.

1.4 Raster files

Raster files (or grid files) are images made of pixels (like digital photos) to which are attached

geographical coordinates. Raster files are convenient formats for supporting large amounts of

data and image processing algorithms used in GIS.

A well known basic raster format is .tif or .tiff which is commonly called GeoTiff when it is used

in GIS. A Geotiff is in fact the association of a .tif file (which can be viewed in any image viewer)

and a small file providing the information on the geographical referencing of the pixels of the

map. Many other raster formats exist. Modern software packages can easily convert maps

from one format to another.

Raster files needed for POPtoGUF are:

• Global Urban Footprint (GUF) : the built up area 2012; the standard distribution at 80 m of

resolution, obtained from the ESA U-TEP platform run by DLR. Note: GUF is accessed via

user agreement with the German Aerospace Agency (DLR). See also, TEP Urban platform

( urban-tep.eo.esa.int /

• Land Cover maps: needed to map areas with dispersed population not sensed by GUF; In

this example we used a publicly available international source: the European Space Agency

Climate Change Initiative (CCI) global land cover at 300 m of resolution is an acceptable

input, easy to download (https://www.esa-landcover-cci.org/). Higher resolution land cover

maps are available in some countries or from regional intuitions such as ICIMOD, often in

vector format.

• Maps of hazards or risks perimeters/areas: needed either as shapefile or in raster format.

For test purposes, global maps can be accesses as raster files via the international data

sharing platforms of UN Environment GRID Portal

7

(http://www.grid.unep.ch/index.php?lang=en and the Group on Earth Observations (GEO)

(www.geoportal.org ).

For the purpose of this manual, data have been

downloaded from the UNEP GRID Global Risk

Data Platform

http://preview.grid.unep.ch/

Also available at: www.geoportal.org

Flood hazard has been retained for the present

example. The flood hazard layers of historical

flood hazard probability occurrence over 100

and 500 years produced for the GAR2015

assessment have been recovered from

http://preview.grid.unep.ch/geoserver/gwc/de

mo/GAR2015:flood_hazard_500_yrp?gridSet=EP

SG:4326&format=image/png � The Darmouth Flood Observatory at University of Colorado (a

contributor to UNEP products) is developing its website with new

products http://floodobservatory.colorado.edu/

2 Upload layers to QGIS and prepare downloaded data

In the step we upload the datasets into GIS and begin to process the data for the layered

analysis. As some uploaded global layers may be very large and cannot be computed easily,

some steps are includes with the purpose of downsizing the files or eliminating unnecessary

data or layers from the project. This includes clipping global maps to the project extent in order

to avoid processing irrelevant data.

2.1 Upload the national boundaries shapefile (ADM0)

Use:

The same button is available on the vertical set, left of the pane: . The ADM0.shp is the map

of the whole country. It will be the extent of the project.

8

2.2 Upload and pre-process GUF Data

This can be done from the Menu simply as:

Layer /Add Layer /Add Raster Layer

Raster files can be supplied as rectangular

tiles representing relevant quadrants of the

globe to avoid downloading and processing

useless data. GUF tiles are named according

to the geographical coordinates of their top-

left and bottom-right corners.

For Thailand, the tiles are:

GUF28_DLR_v01_e095_n10_e100_n05_OGR28.tif










Note: pixel values from the GUF .tif files are 0 and 255 (they will be converted later to 0-1)

2.2.1 Merge GUF tiles into one layer

Use QGIS/Raster/Miscellaneous/Merge

9

Select all input tiles and save the merged output file as GUF28_DLR_THA.tif.

2.2.2 Clip out unused GUF data

The pilot study used in this example produces output for the country of Thailand. Therefore,

unused data are data outside the national (ADM0) boundary.

The QGIS Clipper tool can be used to combine the GUF merged file (GUF28_DLR_THA.tif ) with

ADM0 to delete values out of the boundary. In terms of the project/grid system, GUF, as all

global datasets, is provided in EPSG:4326. Therefore, we can select ADM0 as the mask .shp

layer as it is also in EPSG:4326.

Use QGIS/Raster/Extraction/Clipper

Save as GUF28_DLR_THA_clip.tif

10

2.3 Upload and pre-process land cover and hazard maps

The ESA-CCI land cover map is used in order to assess population not living in buildings/houses

detected by GUF pixels. This file is 2015 map with 300m (grid-size) resolution, version 2.07. In

short-hand the file is: ESACCI-LC-L4-LCCS-Map-300m-P1Y-2015-v2.0.7. The ESA-CCI land cover

map is a single file for the global dataset (in EPSG:4326 coordinate system). We will extract a

work area for our project in order to avoid heavy calculations on irrelevant data.

The ESA-CCI land cover contains values with reference to a classification of land cover types

following the Land Cover Classification System (LCCS) developed by the United Nations Food

and Agriculture Organization (FAO). For more information on this dataset, see the ESCA-CCI

User Guide: http://maps.elie.ucl.ac.be/CCI/viewer/index.php.

For this example pilot study, download the global flood risk map from UN Environment Grid

Portal or GeoPotal. We accessed the flood risk map file called fl1010irmt.tif (from Grid Portal).

It displays flood risks in 5 classes of severity with raster cells of 10x10 km size.

2.3.1 Extract a working window from the global ESACCI land cover and hazad maps

• Zoom ESACCI-LC (and adjust manually the QGIS window size) so that the Map View Extent

matches the ADM0 map.

• Right-click on ESACCI-LC to open “Save As”... Select Map View Extent and save under

ESACCI_LC2015_THA_300m.

Repeat these steps for the flood hazard layer

11

2.3.2 Clip out unused raster data

Use the QGIS Clipper – same as in Step 2.3.2

Save as…in THA_Input folder with a new name, e.g.: ESACCI_LC2015_THA_300m_clip.tif and

FloodriskTHA_UNEP_clip.tif

.

12

Results for the hazard data after initial processing are displayed here in a ramp of greys.

2.4 Save the project

Now we have uploaded and layered the key datasets within our project area and it is a good

moment to save the project (QGIS project file) so that we don’t lose the progress thus far. Use

the Save As... button on left-top. This v0 project will be kept as a link to archives. Save the file

with a short but descriptive name with indication of the version, e.g. Save as THA_v0.

A GIS project is the collection of maps (layers of input files) uploaded and processed, and the

results of calculations, with characteristics such as styles (colours...). A project can be saved at

any moment of the work. When reopening QGIS, you can go directly to state of the last time

that you saved. To avoid problems in the course of the application, it is recommended to

define carefully the project at the very start, which includes extent and projection system.

3 Prepare Shapefiles for Assimilation of Population Data

The following steps apply specifically to this example case study of calculating ‘Pop to GUF’ for

Thailand and these specifications can vary and be adapted for other applications of the

methodology.

3.1 Define project projection system (CRS: Coordinates Reference System)

By default, QGIS opens in “WGS84 EPSG:4326” which is the global system of geographical

coordinates expressed in Arc Degrees, Minutes and Seconds. For calculating values related to

areas, the spherical coordinates need be converted to meters, which means projecting them to

a flat surface. The projection depends from the latitude and longitude of the country (the Earth

is not a perfect sphere!), as well as of the geodic model used; there is no single solution. There

are systems of projection leading to acceptable accuracy of areas measurements by zones, the

most precise ones being defined by countries (and even regions within countries). For the

purpose of the application, the projection system used is called World Mercator (code name:

13

WGS84 EPSG:3395).

3.1.1 Set the CRS for the project

At the bottom-right of the window, click on . You get the first impage below, where

CRS (Coordinates Reference System) is EPSG:4326.

Click on Enable ‘on the fly’ CRS transformation and select WGS84/World Mercator EPSG:3395

in the pane Coordinate reference systems of the world as in the second image below (you may

have to browse the list). Click OK.

1 2

Save the project as THA_v1

Now, when you upload a map to this project (THA_v1), it will be displayed ‘on the fly’ to

EPSG:3395 (the chosen projection system) so that it can be visualized and overlaid with other

layers.

Note: if you are having difficulties with projection, it is possible to force this display [see QGIS

user assistance]. Calculations can be done as well for each file but original files (raster or vector)

will keep their own EPSG. This means that it is not possible to combine 2 layers having different

EPSG (e.g. clipping a raster file with a shapefile, merging two shapefiles ...). Therefore, it is safer

to convert the input layers used for the project into the project EPSG. It can be done by saving

them under a different name and giving them in the saving box the desired EPSG.

In the EPSG3395 project, grid size measurements are in meters.

3.2 Upload the national boundary (ADM0) shapefile to QGIS

3.2.1 Upload the national boundary (ADM0) shapefile to QGIS

Proceed as in Step 2.2., using the button.

3.2.2 Check the CRS of ADM0 in the Properties/General box

If the map is registered in the project’s CRS, there is nothing to do (it may be the case with

14

nationally provided maps...) � Step 4

However, If the map is not registered in the project’s CRS (e.g. because it has been downloaded

from international source such as GDAM and is in WGS84 EPSG:4326, or for any other reason),

we must re-project it to the project’s CRS � Step 3.3

3.3 Project (or re-project) ADM0 to the project CRS

Create a new directory and name it THA_input.

Select the shapefile layer (ADMO) and Save As... a new shapefile named THA_adm0_

EPSG3395 and placed in the new folder THA_Input\ADM.

While saving, change the displayed CRS to the Project’s new CRS (in this case, the project’s CRS

is WGS84/ World Mercator EPSG:3395). “Save as”, but don’t type the file name in the box.

Instead, click on Browse, go the appropriate folder and type the name there (to declare the

full path...).

ADM0.shp

THA_adm0_EPSG33

95.shp

3.3.1 Verify the new coordinate reference system (CRS) for ADMO

Right click (or double click) on the layer’s name to open the Layer Properties box:

and go to General\Coordinate reference system:

15

Remove the old ADM0 file from THA_v1

3.4 Project (or re-project) other Shapefiles (as needed)

3.4.1 Upload the ADM1, ADM2, ADM3 shapefiles to THA_v1

Other national shapefiles for Thailand will be needed for integrating the census data into the

computations. If these files are in geographical coordinates (WGS84 EPSG:4326) or another

CRS, they have to be re-projected into the project’s CRS.

Same method as above, i.e: Save As... change the displayed CRS to the Project’s new CRS (in

this case, the project’s CRS is WGS84/ World Mercator EPSG:3395). “Save as”, but don’t type

the file name in the box. Instead, click on Browse, go the appropriate folder and type the

name there (to declare the full path...).

The destination folder is THA_Input\ADM.

The new files (with CRS WGS84/World Mercator, EPSG:3395) are:

THA_adm1_EPSG3395.shp



4 Create a raster mask layer with properties for data assimilation

Raster calculations involving several layers usually require identical pixels size and alignment.

Some software packages require that raster files have the same extent. QGIS can accommodate

16

various extents in one calculation, except when using the SAGA and GRASS tools, which require

files to have exactly the same extent.

A common data assimilation grid chosen for convenience for raster files is 100m x 100m pixels.

For convenience in computations, the raster files are resampled to this format (100mX100m

grid) when they have smaller (e.g. GUF) or larger (e.g., in this case, land cover, hazard map...)

pixels. This is accomplished with what’s called a mask layer -essentially a reference template

for the project.

For this methodology, we will actually develop two common reference mask raster layers. The

first one will be made of 0 and NoData cells and is used for additions. The second mask layer

contains values of 1 and NoData cells and is used for multiplications. Note: addition to or

multiplication by a cell with value of NoData results in NoData.

These masks will be produced by rasterizing the THA_adm0_ EPSG3395 created at the previous

step. They will be used as exact references when resampling raster files, when rasterizing new

shapefiles, or when clipping out results for Thailand.

4.1 Creation of the 1_No Data Mask with 100m pixels

Go to QGIS Processing Tool Box.

Menu: Processing: Toolbox. Open SAGA/Raster creation tools/Rasterize.

In the command box, select the THA_adm0_EPSG3395.shp vector layer.

Keep all default values for all parameters in the Rasterize function are kept EXCEPT for:

Output Values: change from default ([2] attributes) to [0] data/no data

Fit: change from Nodes to Cells

Cellsize: set to 100 (in case when default value is not 100.00000 as in the example)

Change from “Save to a temporary file” to “Save to file...” and then click , go to the

THA_Input\ folder and give the name THA_Mask_EPSG3395_1_ND.

Notation note: “1_ND” l indicates that this mask has 1 and No Data values.

17

4.2 Creation of the 0_No Data Mask with 100m pixels

Go to QGIS/Processing Tool Box/SAGA/Raster calculus/Raster calculator

Select the new mask raster layer created above (THA_Mask_EPSG3395_1_ND) as the input

layer

Formula: a-1

Save Calculation as: THA_Input/THA_Mask_EPSG3395_0_ND.tif

Save the Project.

18

5 Assimilation of Input data

5.1 Prepare Population Census tables

The required data is population from the census, by administrative regions. Generally, the

lower level of the groupings (e.g. districts or villages) gives more accurate outcome but the

methodology can be implemented at different scales, in this case with data by districts.

For the THA exercise, the Population Census table were downloaded from the CityPop database

(https://www.citypopulation.de/php/thailand-admin.php). From this download, a single file

was created with 2000 and 2010 statistics: CityPop2000_2010_Districts.csv.

Districts codes corresponding to shapefiles used have been inserted in the table. NOTE: This

requires a careful check as common attributes used were names, which sometimes are subject

to spelling variants.

5.2 Introduce Population Census statistics into the project shapefile

5.2.1 Upload CityPop2000_2010_Districts to the project.

If the format is .csv, upload with the “Add vector layer” button

If the format is .xlsx or .ods, upload with the “Add Spreadsheet” button

5.2.2 Join to the administrative region shapefile

For this THA pilot, the census data is associated with administrative region level 2 (districts or

ADM2), i.e: THA_adm2_EPSG3395

• Double-click (or Right-click) on THA_adm2_EPSG3395 and go to Properties.

• Click on Joins and then on the + button [bottom left]

For this example, we utilized a file with population data by districts for 2000 and 2010 and a

common join field called CCA_2:

Join Field: CCA_2 ;

Target Field: CCA_2;

Check “Choose which fields are joined” and select your fields, e.g: CTYP_NAM,

Pop_2000 and Pop_2010 ;

Check “Custom field prefix name” and delete the default text (keep it blank)

Click OK and then OK

Save the file As... THA_adm2_POP_EPSG3395 to a new folder named THA_CALC

19

This file will be used in further steps for calculations.

NOTE: If you forget to save this layer, the “join” will disappear when closing the project.

5.3 (Re)Project and Resample raster layers

The raster layers also need to be projected from WGS84 EPSG:4326 to the CRS WGS84/World

Mercator EPSG:3395 and then to be resampled to the 100m x 100m pixels resolution chosen for

this sample application for Thailand. The files to re-project and then resample are:

GUF28_DLR_THA_clip.tif

ESACCI_LC2015_THA_300m_clip.tif

FloodriskTHA_UNEP_clip.tif

Therefore, conduct step 5.3.2 for each of these raster layers, as needed

5.3.1 Re-project (convert, warp...) each of the raster layers to EPSG:3395

Use QGIS/Raster/Conversion/Warp

Set Input file to be converted

Output files: set (save to) INPUT

DATA\ and the same name as in input

file augmented with EPSG3395

Source SRS: in principle, use default

value, which in this case is EPSG:4326

Target CRS: check and set to

EPSG:3395

Mask layer: use for example

THA_adm2_POP_EPSG_3395

5.3.2 Remove the old (prev. projection) files from the THA_v1 project

Remaining raster files are:

GUF28_DLR_THA_clip3395..tif

ESACCI_LC2015_THA_300m_clip3395.tif

FloodriskTHA_UNEP_clip3395.tif

20

5.4 Resample raster data to common grid

Resampling raster data to 100mx100m grid is done with QGIS/Raster/Raster calculator where

change in “current layer extent” means change in extent AND in pixel size. This function uses

the nearest neighbourhood algorithm. In this case, the the aggrdation/down-scaling does not

noticeably change the distribution of values. Similar resampling can be carried out with the

SAGA toolbox.

5.4.1 Resample GUF data to 100x100m grid

GUF28 pixels have a size of approximately 80m in Thailand. To resample the data with the

nearest neighbourhood algorithm to 100x100m cell size:

Use QGIS/Raster/Raster calculator

Raster calculator expression (formula): "GUF28_DLR_THA_clip3395@1" (introduced by

double clicking the name in Raster bands)

IMPORTANT: In Raster bands, SINGLE click now on THA_Mask_EPSG3395_1_ND.tif AND THEN

click on the “Current layer extent” button. The raster calculator will resample the input map to

the resolution set for Current layer extent (100m), using the nearest Nnighbourhood formula.

Save output layer as THA_Input/GUFDLR_100m.tif

5.4.2 Resample ESACCI land cover data 100x100m grid

Pixels of the downloaded map are of circa 300 m x 300 m. For assimilation in this project, the

ESACCI map is converted to a 100 m x 100 m raster.

21


Raster calculator expression (formula): “ESACCI_LC2015_THA_300m_clip3395@1" (by

Use same procedure as in 5.4.1 and Save output layer as

THA_Input/ESACCI_LC2015_THA_100m.tif

5.4.2.1 Visualize land cover classes

Included with ESACCI land cover raster data for download are metadata for the land cover

classes used and a GIS (.qml) file for integrating the legend and visualization of the classes

(colors) in your GIS package.

22

The procedure will vary depending on the version of QGIS, but generally this step is

accomplished by selecting the layer and going to:

Properties/Style/Load

Select ESACCI-LCMapsColorLegend.qml

5.4.3 Resample Hazard/Risk data (Flood_Risk_UNEP_GRID3395.tif)

Pixels of the downloaded map are of circa 10 km x 10 km. For assimilation in this project, again

the map is converted to a 100 m x 100 m raster.

When the nearest neighbourhood algorithm is used, it does not change the distribution of

values, which are in this case categories (1 to 5).


Use same procedure as in 5.4.1 and Save output layer as

THA_Input/Flood_Risk_UNEP_GRID3395_100m.tif

23

The input file includes No Data pixels and the result is ADM0 rectangular extent, not the ADM0

polygon as previously. So, it is necessary now to Clip out Flood_Risk_UNEP_GRID3395_100m.tif

with ADM0.

Use QGIS/Processing Tool Box/SAGA/Vector<>Raster/Clip Raster with Polygon.

Shapefile: Adm0_EPSG3395

Save Rasterized result as THA_Input/ FloodRiskTHA_UNEPGRID_100m_3395.tif

Results from final processing are displayed here in a ramp of greys.

6 Weighting GUF pixels according to their agglomeration

A basic objective for the Pop to GuF methodology is to downscale population data by utilizing

the GUF, knowing that an assumption of an even density of population per GUF pixel within an

administrative division is not realistic. Obviously, population density is much higher in cities

where land is scarce and houses and building are attached than in villages where land is

24

abundant and individual houses are surrounded by gardens. To capture this uneven density,

POPtoGUF starts from a very simple model where isolated GUF pixels are down-weighted as

compared to GUF pixels in cities (or in areas of agglomerations of built-up areas) which keep

the maximum values in the center and are moderately deflated in the outskirts. This model is

based on Gaussian Filtering or Smoothing. The algorithm is easy to implement in GIS (e.g. with

SAGA) and can be tuned (regarding smoothing radius and intensity) according to the context of

the study. It is a model that can be combined with other models, for example models needed to

assess the dispersed population in habitats that could not be observed by satellite (seet Step 7).

Moreove, POP to GUF can be implemented at various scales, from Regional to National and

local assessments, depending on the detail of population statistics available.

How does Gaussian smoothing (filtering, blurring...) work?

Smoothing is a methodology which transforms crisp data into values taking into account their

neighbourhood. It gives more weight at the central pixels and less weights to the neighbours.

The farther away the neighbours, the smaller the weight. The process is repeated all over the

map (or the image) with a mobile window. In this simplified example of a 5x5 kernel, one cell of

a value of 100 (on a 0-100 scale) is surrounded by cells with zero value. Once smoothed, the

central value drops to 15.02, the total of the whole kernel remaining at 100. If this cell had been

surrounded by cells with values greater than zero, it would have received value in return. In the

case where all neighbouring cell have a 100 value, the result of these exchanges would have

been 100 for the central cell.

�

The program computes (and re-computes successively) the values of all pixels using a mobile

window.

Regarding the issue of weighting GUF pixels according to their agglomeration, the smoothing

methodology keeps the value of towns or agglomerated clusters of urban areas (still in black on

the right hand picture) while it reduces that of small cities and villages (in grades of grey).

0.00 0.00 0.00 0.00 0.00

0.00 0.00 0.00 0.00 0.00

0.00 0.00 100.00 0.00 0.00

0.00 0.00 0.00 0.00 0.00

0.00 0.00 0.00 0.00 0.00

0.37 1.47 2.56 1.47 0.37

1.47 5.86 9.52 5.86 1.47

2.56 9.52 15.02 9.52 2.56

1.47 5.86 9.52 5.86 1.47

0.37 1.47 2.56 1.47 0.37

25

This property is used to weight the average population density (per administrative divisions)

according to the size of human settlements (higher density in large towns than in small cities

and then in villages...). At the end, total population data (net of dispersed rural population) by

administrative division is distributed in proportion to the value of the smoothed pixels. The

total population by administrative divisions is unchanged.

6.1 Clip-out the sea and convert GUF pixels to 1 value

GUF pixels in THA_Input/GUFDLR_100m.tif are given the conventional value of 255 as many

raster files, or 0 (0 meaning no GUF). For calculation purposes, this 255 value will be converted

to 1 (with a Real number format with decimals).

Also, a smoothed pixel exchanges values with its neighbours. This poses a problem on the

shoreline as (in principle) no built-up and related population should be sent to the sea. If

neighbouring cells have 0 values, coastal built-up pixels will lose value to the sea. Fortunately,

the algorithm ignores NoData, so coastal pixels will lose value only on the inland side, not to the

sea if we clip pixels out of THA_ADM0.

Note: This simple solution has a bias for inland borders but because we don’t process data out

of the project area (for this sample application) no correction is done for simplicity reasons. But

an adjustment could be done for transnational projects.

These two operations will be done in one step with the SAGA raster calculator.

Multiply THA_Input/GUFDLR_100m.tif by the 1-No Data raster Mask (created in Step 4.1) and

divide the result by 255 in order to set GUF pixels value at 1.

Use QGIS/Processing Toolbox/SAGA/Raster calculus/Raster calculator

Main input layer: GUFDLR_THA100m.tif [is “a” in the Formula]

Additional layers (optional): THA_Mask_EPSG3395_1_ND [is “b” in the Formula]

Formula: (ifelse(eq(a,255),1,0))*b [It reads: if “a” equals 255, the return 1; if not

return 0; and lastly multiply by “b”.]

26

Save calculated as: THA_Input/GUFDLR_THA100m_clip.tif

Now GUF values have been clipped to the mask layer and pixel values are converted to 0 and 1

(and NoData, outside the project area – e.g. in the sea).

6.2 Smoothing Clipped GUF data

As mentioned above, the parameters (smoothing radius and standard deviation) for the

Gaussian smoothing operation can be adjusted to fit with the expected reality. This is a matter

of calibration and different options may be tested according to a priori knowledge of the way

that populations are distributed across space and urban agglomerations for the specific country

or project area.

For this pilot test case for Thailand, the parameters selected are:

• Smoothing radius: a multiple of the cell size. 5 is a commonly used value and corresponds in

the case of 1 ha cells to a little bit more than 1km2 in the case of a square search and a little

bit less for a circular search. Higher values for the radius can be considered or tested, noting

that since the values are decreasing in proportion to the square of the distance from the

center pixel, they rapidly approach zero anyway.

• Standard deviation (std): it is the measurement of the quantity of value which will be

spread out over the pixel. It is expressed in number of standard deviations in relation to the

chosen radius. The larger the radius, the higher should be the standard deviation to get

meaningful results. For the THA pilot project, with radius of 5, std is 1. This is an empirical

choice, resulting from various experiments in several countries. It can be easily changed in

the SAGA toolbox.

27

Use QGIS/Processing Toolbox/SAGA/Raster filter/Gaussian filter

Input: THA_Input/GUFDLR_100m_clip.tif

Standard deviation: 1

Search mode: Circle

Radius: 5

Calculated result saved as: THA_CALC/GUF_sm1_5.tif

28

THA_CALC/GUF_sm1_5.tif

6.3 Clip out smoothed values outside of GUF cells (optional)

Smoothed values range from 0.0001 to 1.0000.

An effect of the smoothing is that smoothed values have been generated in the grid outside of

the cells with original GUF values (cells previously with 0 value have acquired coefficient values

from smoothing by cells within 5 cell radius). These smoothed values could be potentially

analytical useful for other Gaussian Smoothing applications – e.g. examining the effect of the

smoothing on peripherals areas of urban conglomerations and for studying nexus areas along

the boundaries of urban and rural areas.

For the POPtoGUF methodology, we choose to eliminate these values in order to keep the

native footprint of the map of GUF pixels. These unwanted non-native values are easily

eliminated with a simple trick: multiplying the result of the Gaussian Smoothing

(GUF_sm1_5.tif) by the clipped GUFDLR_100m_clip.tif. Since the non-GUF grid cells in the

unsmoothed input layer (GUFDLR_100m) are equal to zero, and GUF grid cells are equal to 1,

the effect of this multiplication is to convert values outside of GUF cells to zero (multiply by

zero) and maintain the values within the GUF cells only (multiply by one).

Use QGIS/Processing Toolbox/SAGA/Raster calculus/Raster product

Grids : Multiple selections/ GUF_sm1_5.tif , GUFDLR_100m_clip.tif

Run

Save product as: THA_CALC/GGUF_sm1_5.tif

Notional note: We introduce the additional G in GGUF to indicated that the smoothed

values have been clipped to the GUF native area.

29

GUF sm 1_5 GGUF sm 1_5

7 Estimate population living out of GUF pixels

At this stage we have completed all the necessary preparations and calculations for applying a

gridded-population density estimation for Thailand using the GUF, under the assumption that

population density, per district of the country, can be estimated by smoothed values for

agglomerations (and location within the agglomerations) of built-up areas, as identified by GUF.

A core principle for the POP to GUF methodology is simplicity, which ensures that the outputs

are relatively easy to understand and consistent with the population census input data.

30

Another advantage of simplicity in the model is that it more easily replicable and can be used as

a baseline upon which more complex calibrations can be designed or tested (or identified via

machine learning techniques), building on the opportunity of accessing the new high resolution

GUF products.

POP to GUF can be customised to supplement standard national outcomes with tailor-made

applications on areas of specific interest, such as regions with specific risks which can be

covered with higher resolution data and more detailed statistics.

For Pop-to-GUF methodology, there is one such customization or calibration to the modelling

which is necessary, and described in this Step 7, utilizing the ESACCI land cover raster data.

The GUF is calculated based on radar sensors from satellites. Remote sensing of houses and

buildings, even with radar sensors, meets limitations. For example, isolated houses or farms

with thatched roofs or under trees (plantations or forest), tents of nomadic people etc, all may

be missing from the GUF cells. Therefore, the core and crucial supplemental adjustment made

to the baseline assumption in Pop to GUF is to estimate locations of dispersed population in

such situations, living in areas that are not mapped by GUF.

There are multiple possibilities for using land cover data to make an adjustment for dispersed

populations that are hidden from the satellite imagery. For the POP-to-GUF methodology, we

use the simple strategy of assigning an estimated average population density for selected land

cover classed in non-GUF areas of Thailand (e.g. cropland and cropland, mosaic tree cover, and

grass lands). This step is one of the simplest possible approaches based on land cover data

and may be refined over time with additional information about these populations.

For the non-GUF areas, we assume an average density - e.g. 30 inhabitants per km2, or 0.3

inhabitants per hectare. This average density value can be modified according to additional

information available for these populations.

7.1 Define area where dispersed population is expected to live outside of GUF

The area where dispersed population is expected to live is defined from a selection of classes of

the ESACCI land cover maps and a correction to eliminate GUF pixels from this area in order to

avoid double counts.

7.1.1 Selection of classes of the ESACCI land cover maps

The classes selected for non-GUF population estimation in the THA study are in red in the

legend. These are the types of areas that tend to have populations not visible from the GUF

directly.

10 Cropland, rainfed

31

11 Herbaceous cover (cropland)

12 Tree or shrub cover (cropland) 20 Cropland, irrigated or post-flooding

30 Mosaic cropland (>50%) / natural vegetation (tree, shrub, herbaceous

40 Mosaic natural vegetation (tree, shrub, herbaceous cover) (>50%) / cropland

50 Tree cover, broadleaved, evergreen, closed to open (>15%)

60 Tree cover, broadleaved, deciduous, closed to open (>15%) 61 Tree cover, broadleaved, deciduous, closed (>40%)

62 Tree cover, broadleaved, deciduous, open (15-40%)

70 Tree cover, needleleaved, evergreen, closed to open (>15%)

71 Tree cover, needleleaved, evergreen, closed (>40%) 72 Tree cover, needleleaved, evergreen, open (15-40%) 80 Tree cover, needleleaved, deciduous, closed to open (>15%) 81 Tree cover, needleleaved, deciduous, closed (>40%)

82 Tree cover, needleleaved, deciduous, open (15-40%)

90 Tree cover, mixed leaf type (broadleaved and needleleaved)

100 Mosaic tree and shrub (>50%) / herbaceous cover (<50%)

110 Mosaic herbaceous cover (>50%) / tree and shrub (<50%)

120 Shrubland 121 Evergreen shrubland

122 Deciduous shrubland

130 Grassland

140 Lichens and mosses

150 Sparse vegetation (tree, shrub, herbaceous cover) (<15%) 152 Sparse shrub (<15%)

153 Sparse herbaceous cover (<15%) 160 Tree cover, flooded, fresh or brakish water

170 Tree cover, flooded, saline water

180 Shrub or herbaceous cover, flooded, fresh/saline/brakish water

190 Urban areas 200 Bare areas

201 Consolidated bare areas

202 Unconsolidated bare areas 210 Water bodies

220 Permanent snow and ice

To extract the selected classes, use QGIS/Processing Toolbox/SAGA/Raster calculus/Raster

calculator

Formula: (a<41)+(a=100)+(a=110)+(a=120)+(a=130)

Output calculated layer: THA_CALC/AgriPlus_THA

This new output layer is a combined presentation of the relevant land cover classes selected in

the previous step.

32

7.1.2 Elimination of dispersed population overlay (duplication) with GUF pixels

In this step we eliminate pixels where GUF may have incidentally overlapped with the land

cover classes selected in 7.1.1 (these 2 raster files are independent datasets from different

sources that have been resampled to the 100x100 m. grid and its possible that some of the GUF

values overlap with some of the pixels in THA_CALC/AgriPlus_THA). These pixels must be

eliminated to avoid double counting.

Subtract the raster data THA_Input/GUFDLR_THA100m_clip.tif from

THA_CALC/AgriPlus_THA.tif , keeping only values >0.

Use QGIS/Toolbox/SAGA/Raster calaculus/ Raster calculator

Main input layer: AgriPlus_THA.tif (is “a” in the formula)

Additional layer: THA_Input/GUFDLR_THA100m_clip.tif (is “b” in the formula)

Formula: ifelse(eq(a-b,0),0,a)

(it reads: if a-b=0, keep 0, if not keep the a value which is 0 or 1)

Output: THA_CALC/AgriPlusNoGUF_THA.tif

The effect of this arithmetic for the output layer is that overlapping values would be less than

zero, and thus eliminated in the new output layer

AgriPlus_THA.tif AgriPlusNoGUF_THA.tif

33

7.2 Estimation of population in Non-GUF pixels

Multiply the new file AgriPlusNoGUF_THA.tif by 0.3 (default mean density)

Use QGIS or SAGA Raster calculator

Formula: AgriPlusNoGUF_THA.tif *0.3 (in SAGA: a*0.3)

Output: THA_CALC/POPAgriPlusNoGUF_THA.tif

NOTE: in the absence of additional information, at this time we use the same assumption for

2000 and 2010, i.e. 0.3 density of dispersed population in the selected non-GUF areas.

7.3 Extract values from pixels to Administrative Regions (e.g. districts)

Now we can link back with the population census statistics joined to the administrative regions

(THA_adm2_POP_EPSG3395.shp) and use this data to assign portions of the actual population

(according to the census) to our GGUF and non-GUF dispersed population areas identified in

the previous steps, according to the assigned density value (.3 per hectare). First, we sum

across the grid for the GGUF_sm values and for POPAgriPlusNoGUF population (as calculated in

7.2) by polygon (administrative region).

Use QGIS/Toolbox/SAGA/Vector<>Raster/Raster statistics for Polygons

Grids: Multiple selections / GGUF_sm1_5 and POPAgriPlusNoGUF_THA.

Polygons: THA_adm2_POP_EPSG3395

Method: [0] Standard

And then, uncheck all output options, CHECK ONLY SUM

Save as to Temporary file.

34

In order to keep the same file name, check that it’s a temporary file. Once the program is run

and the Temporary fie is checked OK, remove THA_adm2_POP_EPSG3395. Then Save the

Temporary file with the same name THA_adm2_POP_EPSG3395 (overwrite). (This is because

QGIS does not allow for direct overwriting of files by using the same file name).

The new file includes the values for the non-GUF areas for each administrative region (district).

Calculate mean population density by administrative division and GGUF pixels points

In step 6, we have assigned a weight to GUF pixels, in the range of 0.01 to 1.00. This weight is a

way of adjusting population density in GUF pixels in order to take into account the

agglomeration factor.

In step 7, we have calculated in THA_adm2_POP_EPSG3395.shp the SUM of GGUF pixels points

by ADM2 divisions. The field is named GGUF_sm.

In step 5.2, we introduced population data in the attributes table of

THA_adm2_POP_EPSG3395.shp. They are the population census data 2000 and 2010 (fields:

resp. Pop2000 and Pop2010) and our estimation of population living out of GUF pixels carried

out in POPAgriPlusNoGUF_THA.tif (field: POPAgri).

The purpose of step 8 is to calculate for each administrative division the population average

value for GUF weighted pixels. The formula is [Population in GUF pixels by ADM2]/ [SUM of

GGUF pixels points].

7.4 Subtract estimated Non-GUF population from Prior to making this calculation, we firstly need to subtract from the total population in each

administrative region to population living outside of GUF – the dispersed population labelled

POPAgri. We name these new fields PGUF00v0 and PGUF10v0.

Open THA_adm2_POP_EPSG3395.shp Attributes Table by clicking on the top window

pane.

Open the Field Calculator by clicking on the abacus (counting frame) icon (circled in red

above).

35

Population in GUF 2000:

Output field name: PGUF00v0

In the bottom right box, click on Fields and Values to open the fields list.

Double click on Pop2000

Click on the “Expression” “minus” icon

Double click on POPAgri

Click OK

Population in GUF 2010:

Output field name: PGUF10v0

click on Fields and Values

Double click on Pop2010

Click on the “Expression” “minus” icon

Double click on POPAgri

Click OK

7.5 Check results and Fix anomalies

At this point, you may find anomalies - shown as negative values, which are irrational. In the

THA test this happended for 26 cases out of 928 divisions. It shows that the formula used for

estimating non-GUF population does not work in all cases. A probable bias may come from

36

landscape patterns with very large agriculture and grassland areas, which leads to irrational

results when using a fixed coefficient. It could also mean that the default coefficient chosen

(0.3 inhab. per ha.) is to high, in general or in some places.

Ideally, the estimation should be improved on the basis of other sources of information on

rural population. In absence of such information for this test, we will modify the previous

formula used for estimation by introducing the condition that when AgriPlus population

estimation is obvioulsly overestimated, we will replace it by another formula. For the example,

when estimated AgriPlus is more than 34% of the population census statistics for an

administrtative area, we force instead a new coefficient for estimation of Total Population in

non-GUF area = 66% of census statistics. [other similar formulas can also be tested…]

With the double condition formula, in ADM2 divisions with high population density, estimations

depend from agriculture/pasture area while in other divisions with low population density and

large agriculture/pasture area, the estimation depends on statistics only.

This calculation cannot be done in the QGIS vector calculator, but it can done using a

spreadsheet such as MSExcel, OpenOffice or LibreOffice (or other…). MSExcel opens the

attribute table as a .csv file. OpenOffice or LibreOffice open the attribute table as a .dbf file

In either case, use the IF function and parameterise it as such:

• For 2000: Name field (column) POPGUF00

If we use OpenOffice or LibreOffice (.dbf format), add to the name the

caracteristics of the field which are ,N,19,0 � POPGUF00,N,19,0

Formula (use Function IF): =IF(R2>P2*0.34,P2*0.66,P2-R2)

Where R2 is POPAGRI

P2 is POP2000

• For 2010: Name field (column) POPGUF10

If we use OpenOffice or LibreOffice (.dbf format), add to the name the

caracteristics of the field which are ,N,19,0 � POPGUF10,N,19,0

Formula (use Function IF): =IF(R2>Q2*0.34,Q2*0.66,Q2-R2)

Where R2 is POPAGRI

Q2 is POP2000

The output of the IF-function formulats in the spreadshet will is as follows: t if R2 is > than

P2*0.33 (a large value), then POPGUF= P2*0.66, if not, take P2-R2 for POPGUF. Again, the

purpose of this additional calculation is simply to force an alternative coefficient for cases (by

district) where the estimated non-GUF populations are irattionaly large.

Save the results:

• With MSExcel: Save as .csv file (accept the warning).

37

• With OpenOffice or LibreOffice: Save as .dbf file.

Open THA_adm2_POP_EPSG3395.shp in QGIS

• If we used OpenOffice or LibreOffice, nothing more has to be done.

• If we used MSExcel we have still to join the results of the .csv table to the shapefile.

Upload THA_adm2_POP_EPSG3395.csv with the “Add vector layer” button

Go to THA_adm2_POP_EPSG3395.shp and double-click Properties and the Joins. Follow the

procedure described in Step 6.3. Keep only the POPGUF00 and POPGUF10 fields.

7.6 Calculate population density per GGUF point

Open THA_adm2_POP_EPSG3395 Attributes Table and use QGIS Vector Calculator.

Note: the default number format in the box is “Integer” (no decimals). It is better having values

with decimals, so we have to declare “Decimal number (real)”

Formula: POPGUF00 / GGUF_SM and Output field name: DPGGUF00

Recall that POPGUF00 is population by adminsitrative region excluding the estimated non-GUF

(dispersed) population.

Notation note: DPGGUF stands for Density of Population per GGUF point

and POPGUF00 / GGUF_SM and Output field name: DPGGUF10

Save the results .

38

8 Produce final gridded population density maps

This final part of the gridded population estimation is accomplished in two steps:

1/ rasterizing THA_adm2_POP_EPSG3395.shp fields: GGUF points’ population density and

2/ multiplying the raster result by GGUFsm1_5.tif (which are our GGUF coefficients)

8.1 Rasterisation of GGUF points’ population density

Use QGIS/Toolbox/SAGA/Raster creation/Rasterize

Shapes: THA_adm2_POP_EPSG3395.shp

Attribute: Firstly DPOGUF00 and then DPOGUF10

Leave default values except:

Output extent: click and Use layer/canvas extent and then Select extent

THA_Mask_EPSG3395_1_ND.tif

Cellsize: confirmat value is 100.0000

Fit: change to [1] cells

Rasetrized: click , go to the THA_CALC folder and give the name DensPOPGGUF2000 (and

then, DensPOPGGUF2010 for the second file)

39

8.2 Calculation of final POPtoGUF Rasterized Results

Multiply the raster layers obtained in the previous step by GGUFsm1_5.tif and add

POPAgriPlusNoGUF for the final result.

Use QGIS/Toolbox/SAGA/Raster calculus/Raster calculator

For POPtoGUF2000

Main input layer: DensPOPGGUF2000 (it will be “a” in the formula)

Additional layers: GGUF_sm1_5 (it will be “b”)

POPAgriPlusNoGUF (it will be “c”)

OK

Formula: (a*b)+c

Calculated: click , go to the THA_CALC folder and type POPtoGUF2000

For POPtoGUF2010

Main input layer: DensPOPGGUF2010 (it will be “a” in the formula)

Additional layers: GGUF_sm1_5 (it will be “b”)

POPAgriPlusNoGUF (it will be “c”)

OK

Formula: (a*b)+c

Calculated: click , go to the THA_CALC folder and type POPtoGUF2010

These are the final results for the Pop-to-GUF estimation, at this point in gridded (raster)

format.

40

9 Estimation of population exposure to flood hazard areas

Now, in the final step, we integrate our population density estimations with hazard areas to

calculate the population exposure to hazard and reintegrate with the administrative region

shapefile to produce the final statistics and present the results.

9.1 Creation of two risk intensity levels: Low risk and High risk

Recall that the UNEP GRID raster data set has been pre-processed in previous steps (resampled

and clipped to THA extent - recall steps Step 2.3.2 and 5.4) to make it compatible with other

layers. The file that we will use now is FloodRiskTHA_UNEPGRID_100m_3395.tif.

As an additional processing step for analysis, we also now convert the data from 5 classes to 2

classes with the sole purpose of simplifying the analysis:

5 classes of risk intensity as distinguished in the UNEP GRID file, from min (1) to max (5). For

simplicity, we will group them into 2 classes only: Low (1,2) and High (3,4,5).

Use QGIS/Toolbox/SGA/Raster calculus/ Raster calculator

Main input layer: FloodRiskTHA_UNEPGRID_100m_3395.tif

Formula: ifelse(lt(a,3),1,0)

Rasterised: FloodRiskTHA_UNEP1_2.tif

Use QGIS/Toolbox/SGA/Raster calculus/ Raster calculator

Main input layer: FloodRiskTHA_UNEPGRID_100m_3395.tif

Formula: ifelse(gt(a,3),1,0)

Rasterised: FloodRiskTHA_UNEP3_4_5.tif

9.2 Calculation of population in Low risk and High risk exposure areas

Use QGIS/Toolbox/SAGA/Raster calculus/ Raster calculator

For 2000

• Low risk (1_2)

Main input layer: POPtoGUF2000.tif (for 2000) or POPtoGUF2010.tif (for 2010)

Additional layer : FloodRiskTHA_UNEP1_2.tif

Formula: a*b

Calculated outcome: PFloLow00.tif or PFloLow10.tif

41

• High risk (3, 4, 5)

Main input layer: POPtoGUF2000.tif or POPtoGUF2010.tif

Additional layer: FloodRiskTHA_UNEP3-4-5.tif

Formula: a*b

Calculated outcome: PFloHi00.tif or PFloLow10.tif

42

Population 2010 living in areas with low risk of flood. Population 2010 living in areas with high risk of flood.

9.3 Extraction of statistics

The outputs from the previous step are still in raster file format. To produce statistics, e.g.

summaries by administrative regions, we must aggregate the information back to groupings by

regions (polygons or vector file). We are interested only in the Sum (i.e. total estimated number

of individuals within the criteria).

Use QGIS/Toolbox/SAGA/Vector<>Raster/Raster statistics for polygons

Grids: select 4 elements: PFloLow00.tif, PFloHi00.tif, PFloLow10.tif and PFloHi10.tif

Polygons: THA_adm2_POP_EPSG3395

Method: [0] Standard

Uncheck all options, keep only SUM

Statistics: [Save to temporary file]

Run

43

If any problems, process raster layers 1 by 1.

Check RESULTS.shp

Remove THA_adm2_POP_EPSG3395.shp from QGIS

Save RESULTS.shp as THA_adm2_POP_EPSG3395.shp and open the file.

44

Remove RESULTS.shp

9.4 Presentation of results

Population 2010 exposed to high risk of flood, by Districts (ADM2)

Exposure to Hazards Assessment based on ‘POP-to- GUF ... · [Final Draft] Aug 2018 Prepared for Expert Group on Disaster-related Statistics for Asia and the Pacific1 Introduction:

Documents